Source-linked AI summary
Understanding Human Mobility from Twitter
Raja Jurdak, Kun Zhao, Jiajun Liu, Maurice AbouJaoude, Mark Cameron, David Newth
TL;DR
Human mobility research needs alternatives to private, low-resolution data, while Twitter’s representativeness remains uncertain. This paper analyzes Australian geotagged tweets using mobility indicators and comparisons with other technologies, finding that Twitter captures overall mobility features and multiple movement modes. The results support Twitter as a proxy for studying human mobility, within the scope of geotagged tweets.
Problem
Existing mobility studies often use private, low-resolution data, while Twitter’s representativeness is uncertain because of sampling, communication, and location biases.
Method
The study analyzes 7,811,004 geotagged tweets from 156,607 Australian users using displacement, gyration-radius, return-time, entropy, and predictability measures.
Results
Twitter mobility patterns show similar overall features to mobile-phone records, while higher resolution reveals intra-site, metropolitan, and inter-city movement modes.
Takeaways & Limitations
Twitter is a suitable proxy for studying human mobility and can support analysis of movement patterns across spatial scales.
Takeaways & Limitations
The study is limited to geotagged tweets, which account for only a small portion of all tweets.
Abstract
from arXiv · showhide
Understanding human mobility is crucial for a broad range of applications from disease prediction to communication networks. Most efforts on studying human mobility have so far used private and low resolution data, such as call data records. Here, we propose Twitter as a proxy for human mobility, as it relies on publicly available data and provides high resolution positioning when users opt to geotag their tweets with their current location. We analyse a Twitter dataset with more than six million geotagged tweets posted in Australia, and we demonstrate that Twitter can be a reliable source for studying human mobility patterns. Our analysis shows that geotagged tweets can capture rich features of human mobility, such as the diversity of movement orbits among individuals and of movements within and between cities. We also find that short and long-distance movers both spend most of their time in large metropolitan areas, in contrast with intermediate-distance movers movements, reflecting the impact of different modes of travel. Our study provides solid evidence that Twitter can indeed be a useful proxy for tracking and predicting human movement.
Introduction
Human mobility research has relied heavily on private, low-resolution tracking data, while Twitter offers publicly available, high-resolution geotagged locations. The paper examines whether Twitter can represent mobility patterns despite sampling, communication, and location biases.
- Research context: Traditional mobility data from phones, GPS, WiFi, and RFID can reveal dynamics but raises privacy and access concerns.Cellular call records typically provide kilometre-scale resolution because measurements depend on distances between towers.
- Twitter as a proxy: Twitter users can geotag tweets with their current locations, providing publicly available mobility data with positioning resolution down to 10 metres.Twitter also offers a large user population for mobility analysis.
- Open questions: Twitter’s representativeness remains uncertain because users form a potentially biased, younger, Internet-connected sample.The paper identifies sampling bias, communication modality, and location bias as three open issues.
- Study aim: The study analyses 7,811,004 Australian tweets from 156,607 users to compare Twitter mobility patterns with patterns observed through other technologies.It uses displacement and gyration-radius distributions to characterize individual and population-level movement.
Results
Geotagged Twitter reveals heterogeneous, multimodal mobility patterns across individual trips, return behavior, predictability, and movement orbits. These patterns vary systematically with travel distance and reflect distinct urban, intercity, and transport-related movement modes.
- Displacement distribution and Radius of gyration: Displacements span 10m–4000km and are better approximated by multiple functions than by a single power-law or exponential.This heterogeneous distribution indicates multimodality in human mobility patterns.
- Displacement distribution and Radius of gyration: 93.6% of displacements fall below the approximately 50km urban-mobility cutoff, while the distribution separates intra-site, metropolitan, and intercity movement modes.The exponential component dominates short relocations, the stretched-exponential component dominates distances above approximately 100m, and the tail represents longer-distance travel.
- Displacement distribution and Radius of gyration: A double power-law fit for 100m–50km suggests two urban mobility modes associated with short and long movements within cities.The two exponents and their separation point represent distinct power-law regimes.
- First-passage time and visitation: A 24-hour fluctuation in first-passage probability matches return dynamics observed in call data records.This similarity suggests that the return pattern is not substantially altered by the communication medium used as the mobility proxy.
- Predictability of Individual Tweet Locations: For users with at least 100 tweets, real-entropy distributions become bimodal when distinct tweeting locations exceed 20, indicating low-regularity and high-regularity user groups.Predictability distributions show the same bimodality, while full historical information retains greater predictive value as location diversity increases.
- Probability density function P(x,y): Movement orbits become more anisotropic from 4km to 200km but more isotropic from 200km to 1000km, while large-radius movers refocus activity around major cities.The intermediate-distance pattern aligns with road travel, whereas the longer-distance pattern is consistent with air travel and city-centered activity.
Discussion
The study finds that geotagged Twitter data reproduce broad mobility features seen in mobile-phone records while revealing movement-scale and predictability differences. These findings support Twitter as a mobility proxy, but the analysis remains limited to the small fraction of tweets that are geotagged.
- Geotagged Twitter mobility has similar overall features to mobile-phone mobility, supporting Twitter as a proxy for studying human movement.
- Higher-resolution Twitter data distinguish intra-site, metropolitan, and inter-city movement, alongside heterogeneous displacement and gyration-radius distributions.
- Twitter users separate into highly persistent, predictable location tweeters and more diverse, less predictable tweeters.
- These mobility patterns can support movement modelling, including user profiles based on radius of gyration for disease-spread analysis.
- The study is limited to geotagged tweets, which account for only a small portion of all tweets.
Methods
The methods construct user location histories from geotagged Twitter records, cluster nearby coordinates into locations, and quantify mobility and predictability from location sequences. Entropy-based measures and Fano’s inequality are used to characterize location regularity and bound predictability.
- The dataset contains 7,811,004 geotagged tweets from 156,607 Australian users collected between September 2013 and March 2014.
- Each user’s location history is represented as timestamped latitude–longitude observations, enabling analysis of movement sequences.
- DBSCAN with ϵ = 250m and nmin = 1 groups spatially proximate records into single locations and can identify arbitrarily shaped clusters.
- Unconditional entropy uses historical visitation probabilities, while real entropy is estimated with a Lempel-Ziv algorithm searching for repeated sequences.
- Fano’s inequality converts entropy into an upper bound on users’ maximum location predictability.